基于改良信度模型的在轨航天器故障诊断理论与方法研究

Traditional evidence theory has been widely used in many applications including the fault diagnosis due to its advantages in uncertainty modeling, reasoning and decision. When dealing with the fault diagnosis of on-orbit spacecrafts, which has disadvantages such as the insufficiency of the fault data, the ambiguity between different fault patterns, and the frequent occurrences of the unexpected fault patterns, great difficulties are encountered and the performances are not well due to the traditional evidence theory's drawbacks. These drawbacks include its requirements of input including data sufficiency for modeling and the consistency between frames, the drawbacks in representation including the subjective and unstrict modeling of ambiguity, the drawbacks in evidential fusion rules including the low robustness and the inability for unexpected fault patterns. Therefore, this project aims to extend the evidence theory and focuses on the research topics including labeling of fault data based on semi-supervised learning, evidence modeling for ambiguous fault patterns, robust belief fusion based diagnosis methods, belief fusion based diagnosis under inconsistent frames, and the open-world belief model. The key scientific problems encountered in aforementioned researches will be emphatically resolved including the uncertainty representation in the semi-supervised learning of the on-orbit spacecraft fault data, the construction of compound focal element for the fault patterns, the evaluation of the fault frame's fitness, and the unknown proposition's representation and reasoning under the open-world assumption. Our project will try to obtain a batch of original research productions, which can be expected to bring breakthroughs in further enriching and polishing the belief model, and to provide solid theoretic foundations and crucial technical preparations for the successful applications of the new research productions in many aerospace fault diagnosis problems.

传统证据理论因其在不确定性信息表示、融合与决策方面的优势，已在包括故障诊断在内的多个领域成功应用。而在面对在轨航天器故障诊断时，因故障数据不充分、故障模式含混、未知故障频发等不确定性问题，加之其自身在输入层面要求数据充分及框架一致；在表征层面含混性建模方法主观且缺乏严谨性；在融合规则层面鲁棒性差、无法处理不一致及开放框架等缺陷，导致应用效果不佳。因此本项目以拓展现有证据理论为目标，重点研究：故障数据半监督标记、含混故障证据建模、鲁棒与非一致框架信度融合、开放框架信度模型等内容。并重点解决上述研究过程中遇到的主要关键科学问题：在轨航天器故障数据半监督学习不确定性表示、故障复合焦元构造、故障辨识框架适宜性评价及开放框架下未知命题的表示与推理。力争取得一批原创研究成果，以期在丰富和完善现有信度模型的工作中有所突破，并为新成果在各类航天故障诊断问题中得以成功应用，提供坚实的理论基础与关键技术储备。

面向在轨航天器故障诊断中的“小数据”、故障模式含混、“非预案”的特点及其诊断过程的实际需求，本项目研究旨在基于信度模型研究在轨航天器故障诊断机理与方法，从理论和应用层面完善与丰富在轨航天器故障诊断理论与方法，并改进证据理论这一重要的不确定性信息融合工具，拓展其应用范围。通过本项目四年来的研究，取得了如下进展和成果：. 首先在故障证据建模层面，设计实现了基于证据推理的半监督分类与半监督回归方法；设计实现了用户可定制的证据建模方法以及兼顾信息损失和运算效率的证据近似方法；分析了区间数距离以及模糊数类型的选取对于故障证据建模的影响；设计了神经证据建模方法以及基于信任区间的故障特征约简方法。在证据建模基础上设计了一系列利用信度模型的故障诊断方法。故障证据建模的研究成果提升了对故障不确定性的描述能力，为后续融合处理奠定了坚实基础。. 其次，在信度融合规则层面，提出了基于学习的模型化组合方法；分析了证据框架和Bayes框架的关系；研究了多种排序融合方法；设计实现了一系列更为客观的证据距离度量；针对融合的拟问题即证据分解开展了卓有成效的研究；针对不同证据源框架焦元多样性问题，设计了条件证据和证据更新规则。融合规则层面的研究进展丰富了信度融合的手段和视角，能实现对不同信源信息的更充分利用。. 再次，基于开放世界框架问题，针对框架不完整问题设计了改进的截集证据函数建模方法，有效实现由故障数据到隶属度，再到证据函数的建模。针对框架不完整的成因之一，即异常点的问题，设计了基于邻域链的更为精准的异常点检测方法。开放世界框架的研究为突发故障模式的检测和判别奠定了良好的基础。. 最后，针对仿真验证系统开展了一系列的工作，如不确定性建模与推理的软件包开发、嵌入式系统开发以及一系列的故障信息预处理方法。还设计了由电路实物模型、多通道数据采集装置和基于LabVIEW的故障监测软件平台。